Abstract: A memetic algorithm (MA) is proposed combining GA's searching capability for breadth with that for depth so as to solve the multi-item single-level capacitated lot-sizing problems (CLSP) of the master production scheduling system, with the infeasible solutions to be dealt with the methods of capacity adjustment and penalty function. To eliminate overload, a "sequencing first, then reversing" strategy is used to adjust the capacity, i.e., checking if the overload occurs in the whole production process then reverse the process. The executive procedure of memetic algorithm is shown by a flowchart and its computation performance is tested and compared with other existing algorithms. Simulation results verified the feasibility and adaptability of the memetic algorithm proposed.